Spatiotemporal Traffic Matrix Synthesis – Public Review

This paper revisits the problem of traffic generation from first principles: rather than following up on previous work of providing an explanatory model for why observed traffic matrices look the way they do, this paper generates TM instances subject to user specified spatial and temporal constraints. The authors hereby rely on the Maximum Entropy (MaxEnt) framework. Maximum Entropy is a powerful set of statistical methods that are loosely analogous to maximum likelihood estimation which has provided useful guarantees on the solutions for a variety of underlying model assumption. Using the maximum entropy framework this paper shows that much of prior work, which ranges from simple gravity models up to the most complex models that have recently been suggested, can be captured by the proposed framework. Indeed, the paper gives multiple examples which illustrate how previous work can be seen as special cases of the MaxEnt approach proposed in this paper.